Skip to main content

Transport

Messaging

Strategy

NATs vs Rabbit MQ

Can use DB Transaction to record whether an event has been published to a queue.

Means saving two items per tranaction. One to the entity of interest the other to the event audit table. See detailed explanation

Mocking Message Bus for Tests

Checklist​

What checklist do you need to evolve when selecting a transport layer for your dev stack?

Listeners​

Standard signiture for a listener is:

import { Message } from "node-nats-streaming";
import { Subjects, Listener, TicketCreatedEvent } from "@stackmates/common";

export class TicketCreatedListener extends Listener<TicketCreatedEvent> {
  subject: Subjects.TicketCreated = Subjects.TicketCreated;
  queueGroupName = "orders-service";

  onMessage(data: TicketCreatedEvent["data"], msg: Message) {
    // extract data

    // take action to persist

    // if success acknowledge the message
    msg.ack();
  }
}

Questions​

Which message bus design decision — topic partitioning, consumer group strategy, or at-least-once versus exactly-once delivery — has the most downstream impact on system reliability?

  • At what event volume does a message bus become essential versus direct HTTP calls being sufficient?
  • How does real-time event streaming change the architecture for agent-native applications that need to react to on-chain events?
  • Which message bus technology (Kafka, Convex, or SQS) is most appropriate for a DePIN application that needs to process sensor data from thousands of devices?